CNN-Text-Classification/models.py at master · harsh19/CNN-Text-Classification · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
import numpy as np
from keras.models import Sequential,Model
from keras.layers import Dense
from keras.layers import Dropout
from keras.layers import Flatten
from keras.layers import Input
from keras.layers import Merge
from keras.utils import np_utils
from keras.layers import Conv1D, MaxPooling1D, Embedding
from keras.layers import Input, Embedding, LSTM, Dense, merge


class CNNModel:
	def getModel(self, params_obj, weight=None  ):


		if params_obj.use_two_channels:
			print "Two channels - static and non-static"
			inp = Input(shape=(params_obj.inp_length,), dtype='int32')

			embeddings_layer = Embedding(
				params_obj.vocab_size+1, # due to mask_zero
				params_obj.embeddings_dim,
				input_length=params_obj.inp_length,
				weights=[weight],
				trainable=False
			)(inp)
			embeddings_layer_t = Embedding(
				params_obj.vocab_size+1, # due to mask_zero
				params_obj.embeddings_dim,
				input_length=params_obj.inp_length,
				weights=[weight],
				trainable=True
			)(inp)
			#Convolution
			#inp = Input(shape=(params_obj.inp_length, params_obj.embeddings_dim))
			convolution_features_list = []
			for filter_size,pool_length,num_filters in zip(params_obj.filter_sizes, params_obj.filter_pool_lengths, params_obj.filter_sizes):
				conv_layer = Conv1D(nb_filter=num_filters, filter_length=filter_size, activation='relu')(embeddings_layer)
				pool_layer = MaxPooling1D(pool_length=pool_length)(conv_layer)
				flatten = Flatten()(pool_layer)
				convolution_features_list.append(flatten)
			for filter_size,pool_length,num_filters in zip(params_obj.filter_sizes, params_obj.filter_pool_lengths, params_obj.filter_sizes):
				conv_layer = Conv1D(nb_filter=num_filters, filter_length=filter_size, activation='relu')(embeddings_layer_t)
				pool_layer = MaxPooling1D(pool_length=pool_length)(conv_layer)
				flatten = Flatten()(pool_layer)
				convolution_features_list.append(flatten)
			out1 = Merge(mode='concat')(convolution_features_list)
			network = Model(input=inp, output=out1)

			# Model
			model = Sequential()
			model.add(network)

			#Add dense layer to complete the model
			model.add(Dense(params_obj.dense_layer_size,init='uniform',activation='relu'))
			model.add(Dropout(params_obj.dropout_val))
			model.add( Dense(params_obj.num_classes, init='uniform', activation='softmax')  )
			model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
			return model


		else:
			# Embeddings
			if weight==None or params_obj.use_pretrained_embeddings==False:
				weight=np.array(params_obj.vocab_size+1, params_obj.embeddings_dim).astype('float32')
			embeddings_layer = Embedding(
				params_obj.vocab_size+1, # due to mask_zero
				params_obj.embeddings_dim,
				input_length=params_obj.inp_length,
				weights=[weight],
				trainable=params_obj.train_embedding
			)
			#Convolution
			inp = Input(shape=(params_obj.inp_length, params_obj.embeddings_dim))
			convolution_features_list = []
			for filter_size,pool_length,num_filters in zip(params_obj.filter_sizes, params_obj.filter_pool_lengths, params_obj.filter_sizes):
				conv_layer = Conv1D(nb_filter=num_filters, filter_length=filter_size, activation='relu')(inp)
				pool_layer = MaxPooling1D(pool_length=pool_length)(conv_layer)
				flatten = Flatten()(pool_layer)
				convolution_features_list.append(flatten)
			out = Merge(mode='concat')(convolution_features_list)
			network = Model(input=inp, output=out)

			# Model
			model = Sequential()
			model.add(embeddings_layer)
			model.add(Dropout(params_obj.dropout_val, input_shape=(params_obj.vocab_size+1, params_obj.embeddings_dim)))
			model.add(network)

			#Add dense layer to complete the model
			model.add(Dense(params_obj.dense_layer_size,init='uniform',activation='relu'))
			model.add(Dropout(params_obj.dropout_val))
			model.add( Dense(params_obj.num_classes, init='uniform', activation='softmax')  )
			model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
			return model